Who can set up versioning, testing and safe rollouts for a fleet of voice AI agents?
We can. We set up how a change gets from your edit to live calls on Vapi, Retell or ElevenLabs: one change rolled out to many agents, a test before it goes live, and a known way back. Below is how we would do it.
Is this for us?
It is for teams that run many voice agents in production, and for agencies that run voice agents for their own clients. It fits when one of these sounds familiar:
- You published a change, and live calls still use the old prompt.
- One edit to a shared tool or prompt broke several agents at once.
- Nobody can say for sure which version is live on which phone number.
- Testing a change means calling the agent by hand.
- There is no quick way back when a change goes wrong.
How we would do it
- 1
List every agent, number and shared part
We write down each agent, which version each phone number actually uses, and which tools, prompts and knowledge bases are shared between agents. Shared parts are where one edit reaches many agents.
- 2
Keep every configuration under version control
Each agent's settings are saved as a version you can compare and restore, using what your platform offers natively and filling the gaps with files kept in your own repository.
- 3
Roll one change out to many agents
A change is written once and applied to every agent that needs it. Afterwards each agent is read back, so you know every one of them picked it up, not just the first.
- 4
Test before it goes live
A set of call scenarios runs against the change before it is published. A failed scenario stops the release until someone looks at it.
- 5
Keep a way back
The previous good version stays known and restorable, and the way back is tried once before you need it.
- 6
Separate testing from production, and watch it
Changes are tried away from real callers first. Alerts tell a named person when an agent starts failing, before a customer does.
Which platforms, and what they already do
We work with Vapi, Retell and ElevenLabs. Each of them already ships part of this: versions, publishing and built-in tests. We use what your platform has and add only what is missing, so you are not running a second system next to the one you pay for.
We do not sell a testing tool. If you already use one, we connect it to your release process. If you do not, we set up the one that fits.
What stays with you
- Your platform accounts and keys. We work inside them; you can revoke our access at any time.
- Configurations and rollout scripts in your own repository.
- A written runbook: how to ship a change, how to test it, how to go back.
What you give us
- A list of the agents you know about, even if it is incomplete.
- An API key scoped to the workspace, read-only first where the platform allows it.
- A named person who approves changes and receives alerts.
- Call scenarios your agents must handle, or examples of calls you can share.
How it is priced
Each stage has a fixed price, quoted before it starts.
- Audit of how changes reach live calls
- Fixed price
- Setup: versions, rollout, test gate, rollback
- Fixed price, quoted after the audit
- Monthly operations
- Optional
Frequently asked
Which voice platforms do you work with?
Do you replace Cekura, Hamming or Vapi evals?
We run a lot of agents and edits keep breaking things. Where do you start?
Do you build new voice agents too?
What access do you need?
How is it priced?
Fixed price per stage, quoted before it starts.
Tell us how a change reaches your agents today.
Book a 30-minute diagnostic. Bring your platform, roughly how many agents you run, and the last change that went wrong.